Why we never documented eBiz — and what changed

For twenty-five years, eBiz has run without real documentation. Not because we didn’t care — because for a platform as comprehensive as ours and this fluid, it was never feasible. By the time you’d written it, it was wrong. So the knowledge lived where it always lives in systems like ours: in people’s heads, answered one email at a time.

In the past year, AI changed that equation twice over. First it made writing comprehensive documentation possible at all. Then it gave that documentation a second audience.

Meet Eddie the eBiz Agent. He may not be able to ski-jump, but if you ask him a question in plain English, you’ll get a cited answer in seconds.

The problem: a platform too big to document

Rebate management undersells what a buying group runs through eBiz in a typical month. Deals are loaded and renegotiated. Rebates accrue across stepped, linear, and individual-target banding. Forecasting projects where each member and agreement will land against target — which means understanding how the engine extrapolates turnover and what happens when a deal changes mid-year.

Members join and occasionally leave — and a leaving member isn’t a deleted row, it’s a cascade of questions about accrued rebate, allocation entitlement, and cut-off dates. Documents are loaded against agreements. Promotions carry expiry dates. Head office keeps hundreds of merchants up to date with news, terms changes, and deadlines.

Every one of those needs explaining. Documenting it by hand was never realistic: the surface area is enormous, the platform changes weekly, and a small team can’t keep hundreds of pages honest. So for two and a half decades, we didn’t pretend to.

What Eddie does

Over the past year we built our documentation from the ground up — for the first time — as structured reference material, with AI doing the work that made it impossible before: drafting at scale, keeping terminology consistent, spotting gaps, keeping pace with a platform that never stops moving.

We wrote it for two audiences from day one. The first is people: a properly organised documentation site for those who want to browse, follow a how-to guide end to end, or read around a topic before making a decision. The second is Eddie. Documentation that an AI can reliably retrieve and reason over needs the same qualities that serve human readers — clear structure, one concept per page, consistent terminology — so every improvement made for one audience sharpens the other.

Eddie is the interface to that knowledge. Ask “How does a linear rebate work?” or “What happens to a member’s rebate when they leave?” and Eddie searches the documentation, reads the relevant material, and answers directly — citing the exact pages it drew from. It doesn’t make things up; the documentation and the assistant are the same knowledge base.

Questions that used to depend on finding the right person now take one message.

What’s next

Eddie is a proof of concept today, deliberately scoped to documentation — a domain where every answer can be verified and cited.

The next step is already planned: feeding our support ticket history into Eddie’s knowledge base. Years of real questions and real resolutions mean Eddie won’t just know how the platform is designed to work — it’ll know the answers to the questions people actually ask.

Beyond that, we’ll explore guiding users through workflows, explaining calculations, and surfacing insight from the data groups already hold in eBiz.

We’d love to hear which questions you’d ask first. For a demonstration of the platform — or of Eddie — get in touch.